Feature extraction from point clouds for rigid aircraft part inspection using an improved Harris algorithm

Abstract

In the process of feature extraction, geometric information about feature points is based on the nearest 1-ring field points, which are usually the edge points of the face adjacent to the feature line. The high structural complexity of rigid aircraft components leads to recognition of false feature points. Therefore, this paper proposes an improved Harris algorithm designed to improve the accuracy of feature extraction. It combines the discrete curvature and the normal vector in a way that is suitable for use with point clouds. Key to this approach is the use of changes in the gradient rather than changes in the curvature to identify corner, plane, and crease points. Our method accurately eliminates pseudo-feature points that are too close to actual feature points in order to refine potential feature points. We demonstrate the efficiency and robustness of the proposed method by validating it against standard models and actual scan data and by comparing it with other feature detection algorithms.